Fumane cave is a south Alpine site well known for its Middle and Early Upper Palaeolithic sequence with Mousterian (units A11 to A5), Uluzzian (A4, A3) and Aurignacian levels (A2 to D1c) (Broglio et al., 2006, Obradović et al., 2015, Peresani, 2012, Peresani et al., 2016 and Peresani et al., 2008). Within the late Mousterian sequence, the unit A9, explored extensively in the last years, is an ensemble of thin levels and lenses composed of frost-shattered breccia and Aeolian silt, diffused with sands and dark sediments because of intense anthropogenic accumulation. Thousands of knapped stones, micromammal, mammal, bird, bones (Fiore et al., 2016, López-García et al., 2015, Peresani et al., 2011, Romandini et al., 2016 and Romandini et al., 2014) human remains (Benazzi et al., 2014) and dwelling structures with hearths and toss-zones have been brought to light; currently, A9 has produced over 50 structures, mostly hearths, scattered at the cave entrance, with more in the western than in the eastern zone, in proximity to the present-day drip-line. The chronometric position of A9 is provided by only one reliable date among a few other measurements of 47.6−45.0 ky cal BP (Higham et al., 2009 and Peresani et al., 2008).

A9 records the appearance of an exclusively Discoid industry sandwiched between two Levallois cultural units: A10 below and A6 above, with the sterile layer A7 in between (Peresani, 2012). The production system was structured in two reduction sequences, a main one and a secondary one, with the same shared goal of producing short, strong, and sometimes pointed implements, as evidenced by pseudo-Levallois points, backed flakes, and subcircular, quadrangular or triangular flakes (Peresani, 1998). The former system exploited blocks and nodules, whereas the latter, which was simpler and less productive, used flake-cores either originating from by-products (cortical flakes) or introduced directly onto the site. The cores began to yield usable blanks right from the initial steps, with the outline gradually changing from unidirectional to Discoid.

The 2012 archaeological campaign explored the western section of the atrial zone, which was limited to the west by the currently visible section at the entrance to gallery A. In proximity to this section, Unit A9II, a thin anthropogenic level containing lithic artifacts, bone, charcoal, and several combustion structures and overlying the A10 stratigraphic complex was removed. The density of the lithic artifacts in that area is two to three times higher than in the adjacent areas. No particular concentrations or preferential distributions of specific artifact categories were identified during excavation, with the exception of structure A9II_SXLII, a concentration of a dozen flakes in sq. 68 g (Fig. 1). Within the assemblage, which covered a few cm², the peripheral flakes sloped towards the center of the structure; some, especially those to the south, were nearly vertical, while those toward the east sloped to the west. This arrangement seems to be due to post-depositional deformation attributable to freezing-thawing mechanisms, but requires confirmation from soil-micromorphological analyses. The size of the artifacts, comparable with each other, the gray Maiolica flint and their state of conservation seem to indicate that the flakes originated from the same core, which was also recovered during the structure's excavation. Therefore, it was simple to refit the entire concentration to form a single reduction sequence. Among the flakes, small fragments and chunks made of the same grey flint, were found.

Refitting was performed primarily with the traditional method, reconstructing the sequence according to the direct consequential position determined by scars and negatives. However, problems soon arose due to the complexity of the task and the need for a punctual analysis. For these reasons, we tried a different approach obtaining three-dimensional templates of the artifacts, in order to be able of analyze them more easily on a virtual system. As previously established in papers on lithic technology and other disciplines within prehistoric archaeology (Bretzke and Conard, 2012, Clarkson and Hiscock, 2011, Lin et al., 2010 and Richardson et al., 2013), the analysis of three-dimensional models can be used to obtain morphometric data that would be otherwise inaccessible with the traditional method. For lithic refitting, the analysis is focused on analyzing not only the pieces found but also to investigate the three-dimensional gaps produced by absent pieces. This method also allows the analysts to bypass preservation and handling issues through an ease of interaction that, in a space without gravity, removes real-world limitations (Gartski, 2016). To obtain the scans, we used a structured light scanner, Breuckmann SmartScan 3D, developed by AICON 3D systems, already performed with excellent results for prehistoric cave sites and artifacts (Breuckmann et al., 2009, Pastoors and Cantalejo, 2014, Pastoors and Weniger, 2011 and Tusa et al., 2013), including lithic samples (Slizewski & Semal, 2009). Details on the instruments and the procedure used for achieving and elaborating the 3D images are presented in Delpiano et al. (2017), who have determined that the 3D approach is very useful to integrate the whole study (in these cases), but not to replace it entirely.

Thanks to this approach, it has been possible to analyze the chaîne opératoire in its entirety recognizing at least 64 detachments that have been set-up in time and space in a matrix. The reassembled flakes construct almost the original shape of the raw material pebble. Few missing artifacts indicate that the gray Maiolica flint cobble has been carried into the site intact or only partially tested after possible collecting in the streambed located a few hundred meters from the site.

The cobble was originally squared shape with smoothed edges and a pronounced convexity on one surface, which was considered appropriate for the application of the Discoid technology. The reduction sequence took place totally at the site as demonstrated from the artifacts produced during the first decortication stages up to the last phase of exploitation and the discard of the core. All the refitted pieces come from structure A9II_SXLII and we did not find any refitted element from other excavated areas.

The morphometric features of each detachment (measurements, edges, presence of cortex or hinged terminations) and the technological position along the reduction sequence frame a list of technical effects designed for each flake obtained. These include decortication, maintenance of knapping faces convexities, creation of the striking platform, mistake reparations and the achievement of the first choice artifacts. The reduction sequence has been partitioned into ten major stages each one featured from specific objectives and consequences and related to the core face exploited (Fig. 2).

The overall development in the core reduction can be seen from the dissection of the available refitted flakes (Fig. 3). The knapper worked out his task with a reduction in a clockwise sense (with a view from the surface A) alternating the production on the two knapping faces, reaching an almost complete 360̊ turn around the core and along the peripheral edge removing much of cortical surface. However, not every detachment strictly follows the previous one in this sense but many shifts or rotations are present; nevertheless, a global trend can be noticed.

The refitted sequence is the expression of the knapper's decision to arrange the exploitation on two opposite, adjacent and secant faces from the earliest stage: the first blows (Stage 1) were designed to remove the cortex from face A, outline a peripheral edge following the natural block convexities and open a striking platform functional to the face B, exploited in the following stage (2). Therefore, the two faces maintain interchangeable roles. However, in these early stages face A appears much more convex than the almost flat face B.

Framed in this concept of non-hierarchical surfaces, the exploitation often insisted on the same striking platforms, wide and basically flat, that have been obtained by the detachment of large flakes mostly cortical and secant. The reduced preparation of the striking platform (a defining feature of Discoid knapping when compared with Levallois) is expressed by flat or convex butts and the platform left “brute”, avoiding any further preparation. The impact points are systematically located on the proximal convex sides of each scar.

Though after the first phases of core design, the reduction proceeds within a fully Discoid concept, with the peripheral edge that is gradually exploited along its total extension by an apparently opportunistic production of secant flakes obtained through chordal or centripetal blows. However, almost all artifacts played the role of predeterminant products and of technical solutions aimed to maintain the central and peripheral convexities. Face B reveals a preferential exploitation: this side of the core is less convex than face A and was knapped for obtaining almost exclusively core-edge removal products, also with cortical back. This face was in fact intentionally worked along its peripheral edge for increasing the central convexity.

The key concepts remain unvaried across the sequence: blows are always practiced starting from the peripheral edge and never attest the intention to deviate from the exploitation of these knapping faces. Core size reduces but is unchanged in its features, being homothetic in accordance with the Boëda's design (2013). In other words, no evidence of a polyhedral tendency has been detected. When discarded, the core measured 38 × 36 × 22 mm (Fig. 4).

Several knapping accidents repeatedly affected and compromised the progress of the reduction sequence, especially in stages 3, 4, 7 and 9. These occurrences have been observed with more frequency on the same face, expressed in hinged fractures, overshoots or frequent fragmentations due to reiterated blows on improperly prepared zones. This pattern denotes an odd and unexplainable insistence in working these areas and could be related to approximate preparation, poor accuracy and misapplication of the technical skills, if available. As a consequence, the knapper rearranged the core through reparatory flakes, which ablated and removed the errors following the same pattern and direction of the previous detachments.

Finally, it has to be noted that the first choice products were poorly represented in the refitted assemblage. This bias is a consequence of the using these products in the cave or in another site, with these tools subsequently lost. In this case, the three-dimensional approach provides clues for depicting and quantifying these missing artifacts.

Summarizing, flake scar patterns combined with the reduction development seen in the virtual model suggest that knapping was organized via unidirectional exploitation until stage 3. Later, until stage 5, the pattern shifted to bidirectional knapping: normally patterned by means of chordal and centripetal detachments, almost perpendicular to each other, which corresponds to the beginning of the core rotation along its peripheral edge. Finally, as the reduction advances, the multidirectional centripetal and convergent pattern reaches its full expression on both the core faces and knapping becomes more structured, completing the first turn of reduction around the core. At last, the final stages show another round of reduction, which produces the smallest amount of flakes making it hard to track a pattern. This exploitation arrangement creates the final classic Discoid core shape with bifacial exploitation.

The missing flakes are non-uniformly distributed across the reduction sequence: selection increased gradually up to the end in relation to the full achievement of the technical objectives.

The major missing artifacts are estimated to number 14, derived especially from stages 1, 5, 6, 9 and 10. Decortication by-products (stage 1) of the core face A is missing, apart from the first one detached. These missing artifacts, obtained undoubtedly in situ, are four cortical flakes with natural or flat butt; only one of which was large (at least 40 × 30 mm) (Fig. 5a). These flakes may have been exploited as a core-on-flake or as a tool equipped with a sharp edge (18 mm long on the left side).

During stage 4, when decortication temporarily stops and the finishing of some peripheral convexities starts, some profitable artifacts were detached and selected. We identified a semi-cortical core-edge-removal flake issued from face B, short and thick in the proximal zone, although provided with a cutting edge on its right side. It also opened a striking platform that was intensively exploited shortly after, according to the concatenation of the two core faces.

In stage 5, the first fine Discoid products are mostly obtained from face A, which after the (unsuccessful) detachment of a pseudo-Levallois point, was deactivated. Knapping shifted to face B, already decorticated, and produced a fine flake through centripetal percussion (Fig. 5.b). It has an asymmetrical triangular section and an oval and elongated shape predetermined from the detachment of lateral flakes.

Stage 6 entails a few blows to face A following the detachment of a somewhat irregular centripetal flake from face B. In this and especially the following stages, fine artifacts were detached and selected, making the volumetric reconstruction tricky. The finest and most easily traceable is a core-edge removal flake with non-cortical left back and right cutting edge (Fig. 7a).

Stage 9 was targeted to achieve first choice products from face A by means of centripetal detachments, but also produced a large number of errors, then repaired. However, an oval flake was taken for its suitable shape. The last artifact of this stage is a core-edge-removal flake achieved by chordal percussion on face A and then picked up. This flake is regular and 5 mm thick, 28 × 17 mm large and equipped with a convex left cutting (30−40°) edge at least 25 mm long (Fig. 6). It certainly represents the most successful product up to this stage of the sequence.

Finally, stage 10 entails the last two detachments on face B before the core is depleted by the last refitted element, which was discarded. The first artifact (Fig. 7b) is a 35 × 26 mm large core-edge removal flake with sharp (45°) and straight left edge. The right back is convex due to previous detachments. The second flake (Fig. 7c) is cross-oriented to the first and overpasses the core-edge. It has a right knapped back, and a left thin and long (28 mm) edge. For this highly functional trenchant, it was selected.

The productivity rate of the complete sequence remains low. Based essentially on the 3D morphology, it seems that only six or seven gestures might have produced usable blanks: four are finely shaped core-edge removal flakes with knapped backs opposed to sharp and convex cutting edges and two are centripetal oval or quadrangular shaped flakes with single or double thin edge. Finally, a thick cortical centripetal flake seems to be the only blank suitable as a core-on-flake for the secondary reduction sequence. Experimental comparisons on Discoid technology testified to an average of seven pseudo-Levallois points obtained from each bifacially knapped core (Bourguignon et al., 2011 and Brenet et al., 2009); however, these tools did not appear to have been the main objective of A9II_SXLII sequence: only one point was obtained and discarded in situ.

Comparing the productivity of the entire A9 lithic industry made on Maiolica flint to the refitting experiment (Table 1) found that at least 30 products (4 cm length + width) were issued from the reduction sequence of A9_SXLII, while the techno-economic structure of the A9 Maiolica industry records an average of 35 artifacts per core on block (therefore excluding the cores-on-flakes). This data remains analogue even if we consider the fragmented, reused, or recycled flakes (Peresani et al., 2015). The A9_SXLII reduction sequence, however, displays a high rate of corticated products: 70% as compared to about 50% of the whole lithic assemblage. This data is related to the bifacial variant of exploitation and to decisions based on knapping design that led the core to maintain cortical portions until its discard. Overall, the frequency of the techno-typological categories of A9_SXLII suggests that this production is fully comparable to the A9 lithic industry (Table 2). In addition, the A9_SXLII core has been discarded at a degree of reduction on line with the cores from A9 (Fig. 8).

The only discrepancies concern the centripetal flakes, which are quite rare in A9_SXLII, probably a consequence of misshapen core convexities or flaking mistakes created when the knapper wanted to make these blanks: errors (and reparations) are in fact higher than the average in A9. The same holds for the cortical backed flakes: we consider that this data could be affected by the strong incidence of these products on the early stages of decortication of face B. As already mentioned, this surface initially appears much flatter, requiring intense knapping on the peripheral edge for shaping the central convexity. As shown by the evolution of the core shape (Fig. 9), the convexity of face B develops throughout the reduction sequence from the onset to deactivation. In accordance with the concatenated exploitation of the surfaces, the need to raise the peripheral edge can be ascribed to the opportunity to exploit non-hierarchical core faces in a coherently Discoid way by also taking advantage of the variability of the method. The preferential exploitation of face B may have been designed to exploit the block in a more flexible and less constricted way by bifacially knapping it. This type of production appears therefore directly related to the applicability and efficiency of Discoid method.

In conclusion, productivity for the Discoid method is not inherently low but it requires high technological investment aimed to maximize flake production on both faces. The core is modified to yield a recurrent series of first choice artifacts. The reality of this method however, is that it can lead to more mistakes and bad technical choices, which bring down the productivity.